claude-code-routervsdiri
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
claude-code-router
musistudio · Agent harness
OSSMCP
Panel
7.04 spec wins
- Reliability
- 6.3
- Usefulness
- 7.7
- Cost
- 8.3
- Longevity
- 5.8
“Its main job is pointing Claude Code at a model that is not Claude, and it is named after both of them.”
diri
cristicretu · Agent harness
OSS
Panel
6.93 spec wins
- Reliability
- 6.8
- Usefulness
- 6.5
- Cost
- 8.3
- Longevity
- 5.8
“It runs plain shells beside the agents, so a human and four robots can now miss the same deadline in parallel.”
Spec by spec
| Spec | claude-code-router | diri |
|---|---|---|
| Architecture | ||
| Category | Agent harness | Agent harness |
| Runs | local | local |
| Platforms | macos, linux, windows, web | macos, linux |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | Yes | No |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | NoCCR is a routing proxy; the agent CLI behind it does the executing. | Yes |
| Multi-file edits | No | YesEdits come from the agent running in the session; diri supplies the worktree, the status and the persistence. |
| Git operations | No | Yes |
| Browser control | No | No |
| Sandboxed execution | YesA documented docker compose deployment runs the router itself in a container; it does not sandbox the agents that call it. | No |
| Multi-agent orchestration | YesCCR fronts many agent CLIs at once (Claude Code, Codex, Grok CLI, Kimi CLI, Kilo Code, OpenCode, ZCode) but does not itself orchestrate agents against a task. | Yes |
| Headless / CI mode | No | No |
| Models | ||
| Backbone | OpenAI, Anthropic, Gemini, OpenRouter, DeepSeek, SiliconFlow, Moonshot Kimi, Mistral, Z.AI, Bailian | Claude Code, Codex, Cursor, Gemini |
| Bring your own model | Yes | Yes |
| Local models | YesAny OpenAI-compatible endpoint can be added as a custom provider with its own base URL, which covers a locally hosted server; no named Ollama or LM Studio preset is documented. | No |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | $0/mo | $0/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | MIT | Apache-2.0 |
| GitHub stars | 37,511 | 351 |
Which one would each critic pick
| Critic | claude-code-router | diri | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 7.3 | 7.3 | no preference |
| El Crítico | 6.3 | 6.8 | diri — Sessions run on remote hosts as well as locally, and nothing documented describes what a remote run does when the link drops mid-task. |
| El Profesor | 7.0 | 7.3 | diri — Session state lives in a daemon rather than in the user interface: closing the app does not end a run, and restarting the daemon restores the conversations. |
| La Inversora | 6.5 | 6.0 | claude-code-router — The entire product is an arbitrage on the gap between what agents charge and what tokens cost, which makes its lifespan a function of somebody else's pricing page. |
| La Jefa | 6.5 | 7.0 | diri — A signed and notarized universal build delivered through a self-updating Homebrew cask is the first distribution story on this shelf I could hand to sixty people unchanged. |
| El Hacker | 8.8 | 7.0 | claude-code-router — MIT, and any OpenAI-compatible endpoint becomes a provider by writing its base URL into the config, which includes the server on my own machine. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.